Six of my workflows are now live in the n8n template library, and n8n has made me a verified creator.
They handle jobs that agencies, online stores and sales teams do by hand every week. They also follow one pattern, which is the part I would most like to talk about here: the model reads and drafts, and plain code or a person decides what happens next.
Audit and rewrite Shopify product data with Claude, Google Sheets, and Slack
Every Monday it scores each published product out of 100, including whether the JSON-LD on the live page matches Shopify on price and availability. Claude rewrites short descriptions, code rejects any rewrite that adds numbers or claims the product data does not support, and Shopify changes only after someone presses Approve on Slack.
Check scheduled Klaviyo campaign offers and links with Claude, Sheets and Slack
Every hour it looks at campaigns about to send, tests their links and images, and has Claude list the codes, offers and dates in each email. Code checks those against an offers sheet and drops anything Claude quoted that is not in the email. It reports to Slack and never edits the campaign.
Update HubSpot contacts from Gmail auto-replies using Claude and Slack
Every hour it reads Gmail auto-replies and permanent bounces. Claude decides whether a contact has left the company or is out of office, and quotes the line that shows it. Left-company and bounce updates wait for Slack approval, and out-of-office return dates become HubSpot follow-up tasks.
Fix broken YouTube description links and offers with Claude, Google Sheets and Slack
Once a week it checks your newest video descriptions for dead links and for offers that are expired or missing, using an offers sheet as the reference. Claude proposes fixes, code checks them against the sheet, and a description changes only after Slack approval. A fix you decline is not raised again unless the description changes.
Why a code check after the model
When the model has to quote its source, checking it is simple: the quote is either in the email or it is not. In the ClickUp and Klaviyo templates, Claude returns exact quotes and a Code node drops any it cannot find. The Shopify and YouTube templates do the same with rewrites, rejecting anything the product data or the offers sheet does not support. A wrong reading is stopped inside the workflow instead of reaching a client, a CRM or a storefront.
All six are on my creator page:
A question for anyone running AI steps on client or customer data: what do you put between the model and the action? I would like to see how others handle it.
If your team repeats a task like these every week, message me and tell me what it is.